The tDAR integration tool

The bases of tDAR’s integration tool are community-defined general ontologies to which variables from individual data sets are mapped (Kintigh et al. 2018). As part of the NSF project, members of the Southwestern Faunal Working Group defined 20 faunal ontologies including various taphonomic variables, which we used to integrate the 32 data sets in this analysis. We included ontologies for taxon, element, proximal/distal portion, burning intensity, butchering, completeness, gnawing, and weathering, plus a “count” column for recording NISP and a “display” column for recording weight.

Once the data sets were uploaded to tDAR and their relevant variables mapped (by the original analyst or project personnel) to the general ontologies in tDAR, we combined them using tDAR’s integration tool. The tDAR integration tool allows the export of the results of the integration for further analysis and to share with colleagues. Kintigh et al. (2018) provides a thorough overview of the tDAR integration tool; here, we detail how the integration for the SWTP was performed, including specific selections that must be made to reproduce our analysis.

In some data sets each line represented a single element; in others there was a “count” column that can be identified as such in tDAR. When a “count” column was present (Figure 1), we selected the variables that corresponded to NISP (as opposed to MNI). In most cases, this meant selecting “Actual #”, “N”, or something similar.

Figure 1: Screen capture of the tDAR integration tool count column selector. Users have to manually select the appropriate “count” variable, if one exists, to represent NISP.

We then added a “display” column for weight (Figure 2), and selected the appropriate column for every data set in which specimen weight was recorded.

Figure 2: Screen capture of the tDAR integration tool display column selector, showing selected weight variables where available.

We then added integration columns (Figure 3). The tDAR integration tool allows the user to select categories within each ontology to be included in the integration, potentially collapsing lower-order categories into higher-order ones. For our analysis, however, we chose to handle the collapsing within R so as to have better control over collapsed categories. Therefore, when adding integration columns, users of the SWTP should click the “Select values that appear in any column” button in the left hand column. This will select the check boxes for categories represented in any of the integrated datasets.

Figure 3: Screen capture of the tDAR integration tool integration column selector. For integration columns in our analysis, users should click the “Select values that appear in any column” button in the left hand column.

Once all integration columns were selected (Figure 4), we requested the tDAR servers to process the integration. Due to the large amount of data in these datasets, the integration tool timed out and we had to request the integration output from the tDAR staff directly (they happily provided it many, many times). The raw output from our tDAR integration reported in this paper is available here.

Figure 4: Screen capture of the tDAR integration tool showing the count, display, and integration columns.

Collapsing taxon data into artiodactyls, lagomorphs, and turkeys

The SWTP focuses on three taxonomic categories: artiodactyls, lagomorphs, and turkeys. Given that many datasets had identified taxa to more specific levels than order, we developed two new packages for R. The first, called tdar, allows R users to access the tDAR Appplication Programming Interface (API), enabling credentialed users to download datasets on tDAR by their tDAR dataset identifier. Here, we use the tDAR package to automatically download the ontologies in the Archaeological Fauna: US Southwest collection on tDAR, though other users might find the package particularly useful for downloading archived archaeological datasets. Access to the tDAR API is currently by request only; please refer to the README document for the tdar package for information on gaining access.

The second R package, treecats, manipulates tree-like categorical data such as the tDAR ontologies and enables users to collapse nested categories in a comprehensive and reproducible way. While the tDAR integration tool allows for collapsing ontologies, it currently depends on the user selecting the appropriate higher-order check boxes, a process that in our experience was tedious and perhaps too error prone to be sufficiently reproducible. It is possible, however, to save a specific integration within tDAR to run again on the same or different data sets. Here the treecats package parses the Archaeological Fauna: US Southwest ontologies and allows for on-the-fly ontological filtering and collapsing.

Using treecats, we collapsed all species in the taxonomic order Artiodactyla, all the species in the taxonomic order Lagomorpha, and included any avian remains categorized as “large aves” in Meleagris gallopavo. Our goal was to conform to established precedent in southwestern archaeology, while using sufficiently broad taxonomic classes so as to account for analyst error. Although it is somewhat common in the US Southwest to include “large mammal” in analyses involving artiodactyl fauna, we did not include large mammal counts in this analysis because some datasets in our analysis only included identified fauna and not elements from this more general category. Thus the results of our calculations would have been incommensurate across sites if large mammal counts were included. For the bone survivorship analysis, we also collapsed the proximal/distal portion data into “Proximal”, “Distal”, or “Complete”.

Data summary

Table 2 summarizes the data sets, including the counts of samples with valid measurements across each variable reviewed in the Southwest Taphonomic Protocol. Our final dataset includes a total NISP of 329,379 across all taxa, including 41,412 artiodactyl specimens, 98,127 lagomorph specimens, and 29,336 turkey or large bird specimens.

Variable Use and Utility

In developing the original Southwestern Taphonomic Protocol, Clark (2014) drew from global faunal literature and several pilot analyses to identify those variables most likely to be effective in distinguishing different aspects of taphonomic histories in southwestern faunal assemblages. In our implementation we found that some of these variables are more useful simply because they tend to be coded by more analysts. More engagement with taphonomic analyses might encourage zooarchaeologists to expand their coding to include at least some of the other variables used in the protocol. In addition, as we noted earlier in this paper, in some datasets relevant taphonomic data had been coded in a catchall variable that included multiple factors such as butchering, gnawing, and weathering. Coding these variables separately would make them more amenable to taphonomic analysis. The fact that they are already analyzed in these datasets should make it relatively straightforward to disaggregate the catchall variable since variable states have already been identified.

We have realized that, given the relative lack of attention to taphonomic variables, at least in the US Southwest, it is not surprising that bone mineral density (used in our bone survivorship analysis) has emerged over the decades as a dominant variable in taphonomic analyses. The basic data (element) that this variable requires are available in most datasets. Table 4 presents the number of datasets for which there were different types of taphonomic data.

The least available SWTP data involved specific elements, artiodactyl astragali and 3rd phalanx. Sample sizes were often quite small and thus the confidence intervals are generally too large for their fragmentation rates to be useful in distinguishing among sites.

Sample size is also an issue for the bone mineral density and bone survivorship computations. Confidence interval ranges are very broad for this comparative analysis because it requires counts of specific identifiable elements, which especially for artiodactyls are few in number. Most artiodactyl bone in southwestern archaeological sites is shaft fragments. Moreover, because the analysis requires proximal and distal portion information, it was only possible to undertake it for twenty of the thirty-two assemblages. The small sample sizes associated with specific artiodactyl elements likely influenced the p values generated in the bone mineral analysis. The p values are all non-significant except for Sand Canyon and Yellow Jacket, for which the relationship between bone mineral density and bone survivorship is highly significant.

Variable-wise Distinctions among Sites

Burning: For the most part, southwestern faunal assemblages exhibit relatively low frequencies of burning, generally below 20% for all taxonomic categories we examined. The one site in these assemblages that stands out for relatively high burning of artiodactyl and lagomorph bone is Frank’s Ruin, which itself was attacked and burned (Solometo et al. 2017). Overall, the highest burning rates are for artiodactyl bone in general, with lagomorph and turkey less so. The distribution of artiodactyl burned frequencies makes the point that it is the analyst who will determine which sites are too taphonomically different to include in an analysis. Is a 15% burning threshold more appropriate (excluding sites from Albert Porter and above in terms of artiodactyl burning frequency), or should the threshold be above 30% burning (excluding Alameda School and above)?

Butchering: Southwestern fauna rarely show evidence of butchering. In all thirty-two datasets it appears in less than 10% of faunal elements and thus is not a distinguishing variable for the late prehispanic US Southwest.

Gnawing: Gnawing was also recorded on less than 10% of faunal elements, with the exception of the slightly higher frequencies on Sand Canyon and San Antonio de Padua. Thus, it too does not appear to be a distinguishing variable for taphonomic histories in the late prehispanic US Southwest.

Weathering: Weathering of faunal bone is also relatively uncommon, at less than 10% frequency for the three taxonomic categories in most of these assemblages. This is not surprising given the dry southwestern climate. Bryant Ranch and Pena Blanca, however, are clearly outliers with weathering being particularly high for lagomorphs and artiodactyls, and Pena Blanca for turkey as well. San Antonio exhibits a relatively high frequency of weathered artiodactyl bone. The high frequency of weathered turkey bone at Gran Quivira is at odds with the other data and may be an artifact of the context in which they were recovered—largely burials as opposed to midden. As we discuss below, high weathering frequencies for Pena Blanca and San Antonio de Padua are likely an artifact of the two-variable coding scheme used in recording the fauna from this site. Note that we have not calculated weathering for lagomorphs at Bryant Ranch as the sample size is below 25.

Fragmentation intensity: The distribution of fragmentation rates is relatively continuous from 30-50%, again making it up to the analyst to decide what the threshold should be for including or excluding faunal assemblages in an analysis based on their degree of fragmentation. A few sites, however, do stand out in terms of higher fragmentation rates Pena Blanca and San Antonio assemblages exhibit high frequencies of fragmented artiodactyl bone, and the rates for HARP and Kite somewhat lower. Pena Blanca lagomorph bone is also highly fragmented, as is San Antonio turkey bone. Artiodactyl bone is more fragmented than lagomorph bone in general, which is not surprising given the more intense butchering of larger game.

Astragal completeness: As noted above, the wide confidence intervals (due to low sample sizes) make this variable are not terribly helpful, but the data do reinforce the relatively high fragmentation of the Pena Blanca artiodactyl assemblage. No other fragmentation analyses, however, indicate that the Pueblo Colorado faunal assemblage is more fragmented than those from the other sites.

3rd Phalanx completeness: Again, the confidence intervals are wide, but San Antonio does appear to be more fragmented.

Artiodactyl bone survivorship: As noted above, only the results for Sand Canyon and Yellow Jacket yielded significant results when calculating Spearman’s rank correlation between bone survivorship and bone mineral density. It is unlikely that artiodactyl bone survivorship is useful without substantially larger sample sizes of artiodactyl elements.

Interpretation of variable output

The immense value of the diversity of variables in the SWTP is that site assemblages have been taphonomically altered by different variables. Thus focusing on only one or two would miss important taphonomic processes. The data also allow analysts to draw more fine-grained distinctions in taphonomic alteration among sites should they wish to do so by selecting specific thresholds in the frequency of different kinds of alteration beyond which they believe assemblages are too modified, or by investigating whether a particular variable has a significant impact on the interpretative value of an assemblage.

Two of the strong outliers in the weathering portion of the analysis, however, alert us to the importance of understanding the nature of the coding scheme the analyst used before assuming that marked differences among sites are behaviorally meaningful. In the case of Pena Blanca and San Antonio de Padua, weathering was coded across two variables, ‘environmental alteration’ which identified the type of weathering on the bone (root etching, polished, corroded, etc.) and ‘environmental degree,’ which was coded from light to heavy. At both sites there were high frequencies of ‘root etching’ that was ‘light.’ In consulting with the original faunal analyst, Nancy Akins, she is of the opinion that this kind of weathering was likely not recorded as weathering by other faunal analysts. The apparent high frequency of weathering recorded at those two sites, then, is probably the result of the more fine-grained recording of weathering than actual differences in the proportion of weathered bone from those recorded at most other sites.

The analysis of these multiple variables also suggests, however, that some that were in the original taphonomic protocol can be removed for late period southwestern faunal assemblages. Astragal and 3rd phalanx completeness information is redundant with the results of the broader fragmentation analysis, and given the sample size issue with these two variables it does not appear to be necessary to include them. Similarly, the relative lack of evidence for butchering and for gnawing on bone, at least for these kinds of southwestern assemblages, indicates that these two variables would not need to be included either.

Intertaxonomic variation

The intertaxonomic graphs allow for a visual inspection of variation in taphonomic history among taxonomic categories both within and among sites. The burning, butchering, and bone survivorship charts provide a good summary that visually brings together the data considered separately in the previous section of the paper. The graphs allow for the more in-depth study of the differential impact that taphonomic factors may have on the taxonomic groups within an assemblage. For example, an examination of the Burning graph indicates that several assemblages, including the Alameda School Site, Franks Ruin, Homolovi IV, and Pena Blanca, all exhibit much higher frequencies of burning among artiodactyl bone compared to either lagomorph or turkey remains. Based on these findings, it appears that burned artiodactyl remains are largely responsible for the elevated level of burning observed in these assemblages.

The Fragmentation comparative chart is particularly important for thinking about substantive comparative analyses among these faunal assemblages. That graph makes clear that the three taxonomic categories vary in fragmentation rates within sites and that this variation is by no means constant across sites. It is thus possible that differential fragmentation rates within and across sites will affect the output from the calculation of ratio indices, such as the artiodactyl index.